Deep-Learning-Based Pre-Diagnosis Assessment Module for Retinal Photographs: A Multicenter Study.

Area of Science:

• Ophthalmology
• Medical Imaging
• Artificial Intelligence

Background:

• Artificial intelligence (AI) and deep learning (DL) show promise for eye disease detection using retinal photographs.
• Automated pre-diagnosis image assessment is crucial for efficient AI-DL algorithm application in clinical settings.

Purpose of the Study:

• To develop and validate a deep learning (DL)-based pre-diagnosis assessment module for retinal photographs.
• The module targets assessment of image quality, field of view, and laterality of the eye.

Main Methods:

• Utilized 21,348 retinal photographs from 1914 subjects across diverse clinical settings.
• Developed the DL module using two algorithms: EfficientNet-B0 and MobileNet-V2.
• Trained, internally validated, and externally tested the module on datasets from Hong Kong, Singapore, and the UK.

Main Results:

• Achieved high Area Under the Receiver Operating Characteristic Curve (AUROC) values for image quality (0.975-0.999) and laterality (0.985-1.000).
• Demonstrated perfect AUROC (1.000) for field-of-view assessment across all datasets.
• Showcased excellent performance and generalizability across different centers and ethnicities.

Conclusions:

• The developed three-in-one DL module accurately assesses retinal photograph quality, field of view, and laterality.
• This module can enhance AI-based models for improved disease screening and diagnosis workflows.